Urine endogenous metabolic markers and application thereof

Abstract

The invention relates to a group of urine endogenous metabolic markers and application thereof, and the urine endogenous metabolic markers comprise two or more than two metabolic markers of 4-pyridoxine, phosphate PA (27: 6), 5, 6-dihydrothymine and glycodeoxycholic acid. The relative expression level of the marker provided by the invention is closely related to the capecitabine-related hand-foot syndrome, the marker is screened through machine learning, an intersection is taken, a refined and stable Logistic regression prediction model is constructed, the occurrence of the hand-foot syndrome is accurately predicted in advance, and the method has the advantages that the effect of accurately predicting the occurrence of the hand-foot syndrome is achieved while the effectiveness of chemotherapy is ensured. And the method has important significance in improving the chemotherapy safety.

Description

Technical Field

[0001] The present invention belongs to the field of colorectal cancer diagnosis, and particularly relates to a group of urine endogenous metabolic markers and applications thereof. Background Art

[0002] Colorectal cancer (CRC) is the third most common malignant tumor in the world, ranking fourth and third in common cancers in women and men respectively, and ranks second in cancer-related deaths worldwide (Freddie, Bray, Jacques, et al. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. [J]. CA: a cancer journal for clinicians, 2018.). Postoperative adjuvant chemotherapy is one of the effective means of clinical treatment of CRC. The National Comprehensive Cancer Network (NCCN) guidelines recommend capecitabine as a basic chemotherapy drug for the treatment of colorectal cancer. Although postoperative adjuvant chemotherapy can significantly improve the long-term survival rate of cancer patients, chemotherapy can lead to a series of adverse reactions, mainly hand-food syndrome (HFS). The incidence of chemotherapy-induced hand-foot syndrome is as high as 50-70% (Li M, Sun X, Yao H, et al. Genomic methylation variations predict the susceptibility of six chemotherapy related adverse effects and cancer development for Chinese colorectal cancer patients [J]. Toxicology and Applied Pharmacology, 2021, 427: 115657.), ranking among the top three adverse reactions related to capecitabine chemotherapy. The occurrence of HFS not only affects the quality of life of patients, but also limits the benefits of chemotherapy for patients, bringing a heavy burden to both patients and society.

[0003] At present, the mechanism of action of chemotherapy-related HFS is still unclear, and there is a lack of effective preventive measures in clinical practice. In recent years, studies on the prediction of adverse reactions have mainly believed that the occurrence of HFS is related to individual gene polymorphisms and the exposure of capecitabine and its metabolites in vivo, but the conclusion is still unclear (Ezzeldin, HHMethylation of the DPYD promoter: an alternative mechanism for dihydropyrimidine dehydrogenase deficiency in cancer patients. [J]. Clinical Cancer Research An Official Journal of the American Association for Cancer Research, 2005, 11 (24): 8699.).

[0004] Metabolomics is an important part of systems biology and one of the important technical means of precision medicine. Metabolomics can reveal the physiological and pathological state of the body in real time and reflect changes related to the body's phenotype by detecting the levels of small molecule metabolites in body fluids such as blood and urine samples. At present, many metabolic biomarker prediction models have been used for early diagnosis of CRC based on blood, urine, feces and tissue samples.

[0005] However, no study has used urine metabolite levels to predict individual differences in HFS in CRC capecitabine chemotherapy. Therefore, developing highly sensitive and specific integrated biomarkers that can be used to predict the occurrence of hand-foot syndrome in advance is of great significance for improving the safety of chemotherapy while ensuring the effectiveness of chemotherapy. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a group of urine endogenous metabolic markers and their application. The relative expression levels of this group of markers are closely correlated with capecitabine-related hand-foot syndrome. By screening the markers through machine learning and taking the intersection, a refined and robust Logistic regression prediction model is constructed, which achieves early and accurate prediction of the occurrence of hand-foot syndrome, which is of great significance for improving the safety of chemotherapy while ensuring the effectiveness of chemotherapy.

[0007] The present invention provides a group of endogenous urinary metabolic markers, including two or more metabolic markers of 4-pyridoxolactone, phosphate PA (27:6), 5,6-dihydrothymine, and glycodeoxycholic acid.

[0008] Preferably, three or more metabolic markers are included among 4-pyridolactone, phosphate PA (27:6), 5,6-dihydrothymine, and glycodeoxycholic acid.

[0009] More preferably, it includes 4-pyridolactone, phosphate PA (27:6), 5,6-dihydrothymine and glycodeoxycholic acid.

[0010] Furthermore, a multi-factor Logistic regression model is constructed based on the metabolic markers to output prediction results.

[0011] Furthermore, the model is: Logit[P=HFS]=4.526+3.694*[4-pyridolactone]+1.699*[phosphate PA(27:6)]+4.557*[5,6-dihydrothymine]-2.516*[glycodeoxycholic acid], wherein [P=HFS] represents the prediction efficiency of HFS based on the combination of metabolite markers, with a distribution range of 0 to 1, and the cutoff value of [P=HFS] is set to 0.645 (the cutoff value of [P=HFS] is calculated based on the maximum value of the Youden index, Youden index=sensitivity+specificity-1); [4-pyridolactone], [phosphate PA(27:6)], [5,6-dihydrothymine], and [glycodeoxycholic acid] respectively represent the relative contents of the metabolite markers in urine.

[0012] Furthermore, when [P=HFS]≥0.645, it indicates that the person being tested has a risk of developing an adverse reaction to HFS; when [P=HFS]<0.645, it indicates that the person being tested does not have a risk of developing an adverse reaction to HFS.

[0013] The present invention also provides the use of a group of urine endogenous metabolic markers in the preparation of a kit for predicting the adverse reaction of hand-foot syndrome in patients with colorectal cancer after postoperative capecitabine-adjuvant chemotherapy.

[0014] The present invention applies non-targeted urine metabolomics technology, adopts LASSO and RF methods and takes the intersection to screen markers, and constructs a prediction model for capecitabine-related HFS adverse reactions based on multivariate logistic regression. The clinical data Y and the preoperative urine data X of the metabolic group of patients with colorectal cancer who received postoperative capecitabine-based adjuvant chemotherapy were obtained, and the data of the embodiment were preprocessed to obtain the data set Entire Set that can be used for subsequent analysis; the samples were randomly stratified and divided into a training set Training Set and a validation set Validation Set at a ratio of 8:2; in the Entire Set, potential biomarkers related to adverse reactions of HFS were preliminarily identified (differential analysis, PCA, etc.); the potential markers were first screened using univariate Logistic regression in the Training Set; then, based on the data after the preliminary selection, LASSO (least absolute shrinkage and selection operator, LASSO) and Random Forest (Random Forest, RF) machine learning algorithms were used for secondary screening to eliminate redundant features; further, the intersection combination of markers obtained by the LASSO and RF algorithms was taken as the final prediction marker, which was used to construct a multifactor Logistic regression model, output the prediction results, and at the same time, verified in the Validation Set to provide a reference for clinical implementation.

[0015] Beneficial Effects

[0016] 1. The urine sample used in the present invention is easy to obtain and non-invasive during collection; in addition, the discomfort of collecting feces and blood samples is eliminated, the compliance of the tested population is higher, and it is easier to be accepted by the testing staff.

[0017] 2. Compared with the existing adverse reaction marker screening methods, the present invention first corrects the interference of various potential confounding factors, can accurately reflect the relevant effects of postoperative capecitabine-based adjuvant chemotherapy on the occurrence of HFS, and reduces the false positive rate and false negative rate.

[0018] 3. The present invention uses urine samples of colorectal cancer patients before surgery (chemotherapy has not yet started) as research objects. Preoperative samples can better reflect the differences in individual physical functions.

[0019] 4. The present invention discovered for the first time that the metabolite level in urine is closely related to capecitabine-related HFS. The combination of metabolic markers was screened by taking the common intersection of two machine learning algorithms to obtain a more refined marker combination, avoiding the randomness of marker screening by a single algorithm; at the same time, a robust HFS adverse reaction prediction model was constructed in combination with the classic Logistic method, and the model performance was verified. In actual application, chemotherapy HFS can be predicted only by the preoperative urine metabolic biomarkers of colorectal cancer patients, which helps clinicians to screen out patients susceptible to HFS adverse reactions before chemotherapy based on the prediction results combined with other conventional means, and take corresponding preventive measures to reduce the risk of chemotherapy HFS in patients, so as to achieve the purpose of improving the safety of capecitabine chemotherapy, and has clinical promotion value.

[0020] 5. When the cutoff value of [P=HFS] in the metabolic marker combination prediction model provided by the present invention is 0.645, the prediction model has the best prediction efficiency. A variety of indicators are used to evaluate the model (also applicable to data sets with unbalanced grouping). The area under the ROC curve is above 0.9 in the training set and the validation set, respectively. The accuracy, precision, and recall rates are all above 0.8, which has a good prediction effect and the results are accurate and reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The figure is a flow chart of the method for using the markers of the present invention.

[0022] Figure 2 It is a PCA scatter plot of patients with grade 0 HFS, grade 1 HFS, and grade 2-3 HFS in the embodiment of the present invention.

[0023] Figure 3 The ROC curves of the Training Set, Validation Set and Entire Set groups for the prediction of capecitabine-related HFS using the four metabolite markers of the present invention are shown. DETAILED DESCRIPTION

[0024] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0025] Example 1

[0026] The construction of a prediction model for HFS associated with postoperative capecitabine chemotherapy in colorectal cancer, such as Figure 1 As shown:

[0027] Step 1: Establish a data set. Obtain clinical data Y (including the occurrence of HFS adverse reactions and general information of each patient) and non-targeted metabolomics preoperative urine data X of colorectal cancer patients who received postoperative capecitabine-based adjuvant chemotherapy;

[0028] Methods: Clinical data (complete HFS adverse reaction follow-up records, general information: age, height, weight, BMI, tumor location, tumor stage) and preoperative urine data of non-targeted metabolomics of 63 patients with colorectal cancer who underwent radical surgery and received postoperative capecitabine-based adjuvant chemotherapy in Shanghai Changzheng Hospital from January 2018 to April 2019 were collected.

[0029] Specifically,

[0030] Step 1.1: Sample collection

[0031] Collect 15 mL of the patient's mid-morning fasting urine within 1-2 days before surgery in a cryopreservation tube and store it at -80°C for later use.

[0032] After the patients completed the first chemotherapy, chemotherapy follow-up was performed. The clinicians determined the occurrence of HFS based on the Common Criteria for Adverse Events (CTCAE), divided it into grades 0 to 4, and recorded the incidence and severity of HFS. According to the patients' chemotherapy adverse reaction records, a total of 33 cases had grade 1 HFS adverse reactions, and 9 cases had grade 2 or higher HFS adverse reactions.

[0033] Step 1.2: Sample analysis

[0034] Before the experiment, urine samples were taken out and thawed on ice. 100 μL of urine sample was transferred and 400 μL of methanol (containing internal standard: choline-d4, 2.0 μg / mL) was added. After mixing, vortexing was performed for 1 min. After centrifugation at 14000g for 15 min at 4°C, the supernatant was collected and stored at -20°C for subsequent LC-MS / MS analysis. QC samples were prepared by taking an equal volume (10 μL) of each study sample and mixing them evenly.

[0035] The detection conditions of the LC-MS / MS described above are: using UHPLC-Q-Exactive-MS technology, the chromatograph is an Agilent 1290 UHPLC system (Agilent Technologies, Inc., USA), and the mass spectrometer is a Q Exactive mass spectrometer (Thermo Fisher Scientific, Inc., USA); chromatographic column: ACQUITY UPLC BEH C8 chromatographic column (1.7 μm, 2.1×100 mm, Waters, USA) is used in positive ion mode, and ACQUITY HSS T3 chromatographic column (1.8 μm, 2.1×100 mm, Waters, USA) is used in negative ion mode; column temperature: 50°C; flow rate: 0.3 mL / min; injection volume: 10 μL; mobile phase A is 0.1% formic acid water, and mobile phase B is acetonitrile (containing 0.1% formic acid). The gradient elution mode was adopted, and the settings were as follows: 0-1min, 5% B in positive ion mode; 1-24min, 5-100% B; 24.1-27.5min, 100% B; 27.6-30min, 100-5% B; 0-1min, 5% B in negative ion mode; 1-18min, 5-100% B; 18.1-22min, 100% B; 22.1-25min, 100-5% B. The injection volume was 10μL, and all samples were injected in a random order. In order to eliminate the potential impact of long-term operation, environmental changes and other factors on the detection results of the mass spectrometer and ensure the stability of the system, when testing the research samples, one QC sample was run in the middle of every 10 samples, and an elution program was added after every 10 samples were run.

[0036] Step 2: Data preprocessing. The metabolomics data in the data set were subjected to internal standard correction (the internal standard was choline-d4), 80% rule processing (retaining characteristic ion peaks with a frequency greater than 80% in any group), QC quality control (eliminating metabolic features with a coefficient of variation CV>0.4 in QC samples), normalization processing (log10 conversion, pareto scaling), compound identification, data outlier detection (IQR method to detect outliers in each group) and filling (KNN-sample wise), and confounding factor correction (using the removeBatchEffect function in the limma package to eliminate the following confounding factor effects: age, height, weight, BMI, tumor site, tumor stage), and then the Entire Set data set for the preliminary exploration stage was obtained; then, the Entire Set was randomly stratified and split into the training set and the validation set according to the ratio of 8:2.

[0037] Step 3: Identification of differential metabolites related to HFS adverse reactions. Urine samples collected from patients 1-2 weeks before surgery were selected to predict whether HFS adverse reactions would occur. First, in the Entire Set, a differential analysis was performed to analyze the differences in urine metabolic levels between the HFS group (grade 1, grade 2-3) and the HFS group (grade 0). Capecitabine HFS-related metabolites were screened by integrating the results of a total of 3 pairwise comparison group differential analyses, that is, at least in any pairwise comparison of any group (grade 1 HFS vs grade 0, grade 2-3 HFS vs grade 0, grade 2-3 HFS vs grade 1), the following thresholds were met: P<0.05 and |log10FC|>0 (using the limma package for differential analysis). Further, exogenous metabolites were eliminated. Principal component analysis (PCA) was used to analyze the inter-group differences in each HFS grouping, and the results showed that ( Figure 2 ), these differential endogenous metabolites can clearly distinguish patients with different severity levels of HFS.

[0038] Step 4: Preliminary screening of potential markers related to adverse reactions of HFS. In the Training Set, the differential metabolites obtained by the above screening were analyzed by univariate logistic regression to screen out potential markers related to adverse reactions of HFS (P<0.05).

[0039] Step 5: Final determination of candidate potential markers related to adverse reactions of HFS. The metabolites obtained in Step 4 were further screened by two machine learning algorithms, LASSO and RF, and the top 10 most important potential markers of each algorithm were screened out.

[0040] Among them, the LASSO algorithm was based on the "non-zero regression coefficient" through 10-fold cross validation; the random forest generated 1,000 trees and selected the top N most important candidate markers of each algorithm based on the "Mean Decrease Accuracy" indicator, where the value of N was 10.

[0041] To further determine the optimal marker combination, the intersection of the top 10 features of LASSO and RF was selected, and the candidate markers were screened twice, and finally 4 candidate predictive markers were obtained (Table 1).

[0042] Table 1 Potential predictive metabolic markers for adverse reactions of hand-foot syndrome

[0043]

[0044] The results showed that the relative contents of the above four predictive metabolic markers had significant differences and important biological effects between the HFS group (grade 1, 2-3) and the non-HFS group (grade 0).

[0045] Step 6: Construction and evaluation of prediction model.

[0046] Further, the final model was determined: a multi-factor Logistic regression model was constructed for the 4 intersection candidate predictive markers obtained from the above screening:

[0047] Logit[P=HFS]=4.526+3.694*[4-pyridolactone]+1.699*[phosphate PA(27:6)]+4.557*[5,6-dihydrothymine]-2.516*[glycodeoxycholic acid].

[0048] The [P=HFS] value was calculated, where the cutoff value of [P=HFS] was set to 0.645 (the cutoff value of [P=HFS] was calculated based on the maximum value of the Youden index). When [P=HFS]≥0.645, it indicated that the person being tested had a risk of developing an adverse reaction of HFS; when [P=HFS]<0.645, it indicated that the person being tested did not have a risk of developing an adverse reaction of HFS.

[0049] Then, the relative content data of the above four urine metabolic markers obtained in the Validation Set are input into the multi-factor Logistic regression prediction model constructed based on the Training Set, and the model will automatically output the prediction results.

[0050] Finally, the prediction performance of the model was evaluated in the Training Set, Validation Set, and Entire Set using indicators such as area under the ROC curve, accuracy, precision, and recall. Figure 3 As shown in Table 2, the areas under the ROC curves of the model constructed in the present invention in the Training Set, Validation Set, and Entire Set were 0.953, 0.955, and 0.954, respectively, and the precision and recall rates were both >0.9. Therefore, the prediction model of the present invention can successfully distinguish between patients with HFS and those without HFS, thereby accurately predicting the risk of HFS in advance, and the results are reliable and have clinical promotion value.

[0051] Table 2 Model performance of the hand-foot syndrome adverse reaction prediction model

[0052]

Claims

1. A set of urine endogenous metabolic markers, characterized by: Including two or more metabolic markers among 4-pyridolactone, phosphate PA (27:6), 5,6-dihydrothymine, and glycodeoxycholic acid.

2. The set of urine endogenous metabolic markers according to claim 1, characterized in that: Including three or more metabolic markers among 4-pyridolactone, phosphate PA (27:6), 5,6-dihydrothymine, and glycodeoxycholic acid.

3. The set of urine endogenous metabolic markers according to claim 1, characterized in that: Including 4-pyridolactone, phosphate PA (27:6), 5,6-dihydrothymine and glycodeoxycholic acid.

4. The set of urine endogenous metabolic markers according to claim 1, characterized in that: A multi-factor Logistic regression model is constructed based on the metabolic markers, and a prediction result is output.

5. The set of urine endogenous metabolic markers according to claim 4, characterized in that: The model is: Logit[P=HFS]=4.526+3.694*[4-pyridolactone]+1.699*[phosphate PA(27:6)]+4.557*[5,6-dihydrothymine]-2.516*[glycodeoxycholic acid], wherein [P=HFS] represents the prediction efficiency of HFS based on the combination of metabolite markers, with a distribution range of 0 to 1, and the cutoff value of [P=HFS] is set to 0.645; [4-pyridolactone], [phosphate PA(27:6)], [5,6-dihydrothymine], and [glycodeoxycholic acid] respectively represent the relative contents of the metabolite markers in urine.

6. The set of urine endogenous metabolic markers according to claim 5, characterized in that: When [P=HFS]≥0.645, it indicates that the person being tested has a risk of developing an adverse HFS reaction; when [P=HFS]<0.645, it indicates that the person being tested does not have a risk of developing an adverse HFS reaction.

7. Use of a group of urinary endogenous metabolic markers as claimed in claim 1 in the preparation of a kit for predicting the adverse reaction of hand-foot syndrome in patients with colorectal cancer after postoperative capecitabine adjuvant chemotherapy.

Images